Knowledge Engineering

Presentation on theme: "Knowledge Engineering"— Presentation transcript:

1 Knowledge Engineering
Develop a Personalized Service Platform With Automatic Customer Catagorization Capability To Enhance Customer Satisfaction And Loyal Customer Retention Good morning ladies and gentleman, It is my honor to introduce my guided reading article entitled ~ 陳孟鈺 、 吳昌儒 National Tsing Hua University (NTHU), Industrial Engineering & Engineering Management (IEEM), Taiwan

2 Outline Introduction Research Objectives
Background Current Services Model Research Objectives System Framework and Customer Categorization Method Case Example and Experiment of Customer Analysis Conclusion 2

3 Background (1/1) 餐飲服務業競爭激烈 客戶忠誠度低 餐飲資訊取得迅速 業界競爭激烈顧客選擇多 業者不易掌握顧客消費習性
服務無法滿足顧客需求 3

4 Current Services Model(1/1)
客戶 1. 我要點啥菜 2. 今天情人節ㄝ 3. 這個服務生好笨 服務生 1. 要幫忙推薦嗎? 2. 今天哪道菜不錯? 3. 找最貴那道好了 訂位服務 迎賓帶位 點菜服務 點菜服務 買單送客 餐前服務 飲料服務 餐中服務 上菜服務 餐後服務 4

5 Outline Introduction Research Objectives
System Framework and Customer Categorization Method Case Example and Experiment of Customer Analysis Conclusion 5

6 Research Objectives (1/1)
建構智慧型顧客服務平台 顧客過去消費記錄以及其個人資料之收集 提供服務人員辨識顧客並提供顧客服務 提供顧客高效率之客製化服務 提高顧客服務滿意度 依顧客之過去歷史消費記錄，使用類神經分類模組進行顧客之喜好進行顧客分類，然後給予套餐之推薦，以提高顧客之服務滿意度 6

7 Outline Introduction Research Objectives
System Framework and Customer Categorization Method Functional modules of the platform Automatic customer categorization Case Example and Experiment of Customer Analysis Conclusion 7

8 Functional modules of the platform(1/4)
System framework 8

9 Functional modules of the platform(2/4)
Customer Registration Module 9

10 Functional modules of the platform(3/4)
Customer Recognition Module 10

11 Functional modules of the platform(4/4)
Customer Service and Ordering Module 11

12 Automatic customer categorization(1/3)
Method 1 Input layer 菜餚推薦清單 1. 影系列套餐 2. 一般套餐 3. 特選套餐 4. 素食套餐 Hidden layer 性別 個性 職業 月收入 消費金額 消費頻率 各餐點消費頻率 Output layer X 12

13 Automatic customer categorization(2/3)
Method 2 顧客重要性 累積消費頻率 累積消費金額 顧客重要性指標 13

14 Automatic customer categorization(3/3)
Input layer 菜餚推薦清單 1. 影系列套餐 2. 一般套餐 3. 特選套餐 4. 素食套餐 Hidden layer 性別 個性 職業 月收入 顧客重要性指標 各餐點消費頻率 Output layer X 14

15 Outline Introduction Research Objectives
System Framework and Customer Categorization Method Case Example and Experiment of Customer Analysis Case discussion Construct and train the BPN model Conclusion 15

16 Case discussion(1/1) 來源:台北某高級日本料理餐廳 資料: 顧客基本資料 顧客消費記錄 性別 職業 個性 月收入
累積消費金額 個人累積點餐次數 個人各餐點點餐次數 16

17 Construct and train the BPN model(1/9)
Attribute Value Gender 0: Female, 1: Male Occupation 1: Student, 2: Government Service, 3: Financial Services, 4: Technology Industry 5: Others Monthly income (NTD) 1: Under $10,000, 2: $10,001 to 50,000, 3: $50,001 to 100,000, 4: Above $100,001 Cumulative expenditure 1: Under 50,000 2: $50,001 to 100,000 4: $100,001 to 150,000 5: Above $150,001 Personality type 1: Quite, 2: Normal, 3: Assertive and outspoken Preference (outcome) A: First Class, B: Economic Class, C: Vegetarian Food, D: Seasonal Specialty 17

18 Construct and train the BPN model(2/9)
Confusion matrix System inferred Customer classified into category i Customer classified into other categories Actual Customer in category i a b Customer in other categories c d 18

19 Construct and train the BPN model(3/9)
Method 1 Model Parameters 1 2 3 4 5 6 7 8 Training epochs 10 150 40 51 48 47 Learning rate 0.3 0.35 0.26 Momentum 0.2 19

20 Construct and train the BPN model(4/9)
Method 1 20

21 Construct and train the BPN model(5/9)
Method 1 Model 8 Predicted value Actual value A B C D Precision Recall 122 10 15 0.782 0.83 18 112 14 0.783 0.778 16 21 150 6 0.824 0.777 3 447 0.987 0.993 Average 0.89 21

22 Construct and train the BPN model(6/9)
Method 2 Model Parameters 1 2 3 4 5 6 7 8 Training time 10 150 40 51 48 47 46 Learning rate 0.3 0.35 Momentum 0.2 22

23 Construct and train the BPN model(7/9)
Method 2 23

24 Construct and train the BPN model(8/9)
Method 2 Model 6 Predicted value Actual value A B C D Precision Recall 126 8 0.829 0.887 11 113 22 0.85 0.774 15 12 162 9 0.835 0.818 2 446 0.98 0.996 Average 0.91 24

25 Method Precision Recall
Construct and train the BPN model(9/9) Method Precision Recall 1 0.89 2 0.91 3-1 0.84 3-2 0.9 25

26 Outline Introduction Research Objectives
System Framework and Customer Categorization Method Case Example and Experiment of Customer Analysis Conclusion 26

27 Conclusion 類神經分類模式結論 類神經模組未來發展 27

28 NO Q &A , and thank you for your listening.
The End NO Q &A , and thank you for your listening. 28